Neurons play tennis

A group of researchers from Australia and the UK was able to teach a small mass of human brain cells to play the first-generation video game Pong. The single-player version of the game consists of a racket and a ball. The player moves the racket along the trajectory of the ball to keep it in play when it bounces off it, as in a real table tennis match.

The cell mass, which the researchers called DishBrain, was grown from adult stem cells. Neural networks in vitro were integrated in silico with a computer using a high-density multielectrode matrix. With the help of electrophysiological stimulation and recording, DishBrain was embedded in a simulated game world simulating a Pong game. Electrical signals were sent to the array to tell them where the ball was. For example, if the electrodes on the right are triggered, the brain cells "know" that the ball is to their left. The frequency of the signal gives the cells information about the distance. Just like in a real video game, the racket can only move left and right, and the goal is to place the racket in front of the ball.

DishBrain, as a person, was trained to play by repeating the game several times so that the system could move the racket and hit the ball. The feedback was electrical signals from neurons.

The researchers found that the system was able to learn how to play Pong in about five minutes – much faster than artificial intelligence. However, they note that DishBrain's skill level was much lower than that of humans or artificial intelligence systems.

The authors hope that this work can lead to improvements in the design of machine learning systems or will be used to test treatments for brain diseases.

Article by B.J.Kagan et al. In vitro neurons learn and exhibit sentiment when embedded in a simulated game-world is published on the biorxiv website.

Aminat Adzhieva, portal "Eternal Youth" http://vechnayamolodost.ru .